A Update on Electron Isolation - WebHome < UCSDTier2 Update on Electron Isolation ... •Bug fix in Secondary Electron Algorithm ... DR < 0.3, w/out bug fix DR < 0.3, with bug fix

A Update on Electron Isolation

Matthew LeBourgeoisVivek Sharma

UCSD Group Meeting5th February, 2008

p.2Matthew LeBourgeois, UCSD Feb. 05, 2008

Table of Contents• To date, have shown muon isolation studies

– http://indico.cern.ch/getFile.py/access?contribId=8&resId=0&materialId=slides&confId=20377

• As well as preliminary electron isolation studies:– http://indico.cern.ch/getFile.py/access?contribId=1&resId=0&materialId=slides&confId=18242

• Today, updates on electron isolation studies– Tracker Based Isolation

• Updated data samples– comparison to previous results– comparison to muons

• Bug fix in Secondary Electron Algorithm– Calorimetry Based Isolation

• Updated data samples– comparison to previous results– comparison to muons

• New “Jurassic Method”– Summary and Recommendations


Electron Types Considered and Data Samples

• For this talk, define different “types” of electrons– Standard GSF Electrons (“pixelMatchedGsfElectrons”)– Previously showed electrons with “medium” PTDR selection

– Electrons with new “loose” Cut-Based (CB) Selection• (Endcap,Barrel)x(bremming w/ E/p ~ 1, low brem, bad track E/p != 1)

Samples are all CSA ‘07simulated with 1_5_2

Ran with CMSSW_1_6_7

Drell-Yan (DY)/DrellYan_ll_40/CMSSW_1_5_2-CSA07-2178/GEN-SIM-DIGI-RECO (800K events)

QCD - Weighted by LO cross section/QCD_Pt_30_50/CMSSW_1_5_2-CSA07-2048/GEN-SIM-DIGI-RECO (1.1M events)/QCD_Pt_50_80/CMSSW_1_5_2-CSA07-2049/GEN-SIM-DIGI-RECO (912K events)/QCD_Pt_80_120/CMSSW_1_5_2-CSA07-2027/GEN-SIM-DIGI-RECO (1.2M events)/QCD_Pt_120_170/CMSSW_1_5_2-CSA07-2171/GEN-SIM-DIGI-RECO (1.2M events)/QCD_Pt_170_230/CMSSW_1_5_2-CSA07-2069/GEN-SIM-DIGI-RECO (931K events)/QCD_Pt_230_300/CMSSW_1_5_2-CSA07-2050/GEN-SIM-DIGI-RECO (784K events)/QCD_Pt_300_380/CMSSW_1_5_2-CSA07-2061/GEN-SIM-DIGI-RECO (1.2M events)/QCD_Pt_380_470/CMSSW_1_5_2-CSA07-2172/GEN-SIM-DIGI-RECO (1.2M events)/QCD_Pt_470_600/CMSSW_1_5_2-CSA07-2096/GEN-SIM-DIGI-RECO (1.3M events)/QCD_Pt_600_800/CMSSW_1_5_2-CSA07-2097/GEN-SIM-DIGI-RECO (602K events)/QCD_Pt_800_1000/CMSSW_1_5_2-CSA07-2028/GEN-SIM-DIGI-RECO (729K events)

Will use CB LooseSelection for this Talk


Electron PT Distributions in DY Sample• Will look at electron candidates with PT > 15 GeV for this isolation study

– In DY, there is an excess of electrons not from the Z/γ*– Did not filter Z →ττ– As well as contaminations from fakes– Bug fix in EID (fixed in CutBasedElectronID.cc from tag 1.8 → 1.9 )

• Contamination in QCD went from 81% to 73%• Selection Efficiency for Z→ee went from 97.3% to 96.9%

All ElectronsCB LooseMC MatchedNot Matched

All ElectronsCB LooseMC Matched


Tracker BasedIsolation


Track Quality Cuts• Previous electron tracker

isolation did not requirequality cuts on the GSF Track

• However, GSF Tracks havefairly similar distributions tothe CTF tracks

Could cut harder on Δdz and d0, but preferto leave loose for early data scenarios

• PTele > 15 GeV

• GSF Track Requirements– d0 < 1mm– χ2 / dof < 10– HitsTracker >= 8

• CTF Track Requirements– PT > 1 GeV– d0 < 1 mm– χ2 / dof < 10– HitsTracker >= 8– Δdz < 5 mm

CTF Tracks for Z→µµ

GSF Tracks for Z→ee


Dealing with the electron footprint1. Inner cone method: Sum all tracks within a ΔR cone around the

electron candidate, while vetoing cones within a smaller ΔR cone.2. Veto candidate track: Sum all tracks within a ΔR cone around the

electron candidate, while vetoing the track that belongs to the electroncandidate.

3. Veto all electron tracks: Sum all tracks within a ΔR cone around theelectron candidate, while vetoing all tracks belonging to electrons.

1. 2. 3.


Update: Extra Electron Removal Method• Algorithmic mistake

in previouspresentation whichled to incorrectextra electronmatching.– Incorrect loop

variable, was notmatching againstextra electrons

– New plots showmore extraelectrons withincone for the DYsample

CB Loose

Before Bug FIx

Before Bug Fix After Bug Fix

After Bug Fix


Quantifying extra electron’s in cone

• Want to know where all the extra electrons are coming from in the DYsample, and whether or not we want to remove them.

• A majority come from Z’s, i.e. are the other leg of the DY decay.– Want to remove these anyway!

0.015 < ΔR < 0.3 23: Z 11: Electron

Overflow:421: D0

1114: Δ-

521: B+

2214: Δ+


ΔR of closest track as isolation variable

• Plot DR of closest track to the electron– Again, add track quality cuts, to ensure track is coming from the same vertex.– Remove matched electron track and all other tracks which match to electrons

as in previous slides– Not as powerful as ΣPT

QCD 50-80DY

Efficiency w/ cut on ΔR > 0.08:DY: 97%QCD 50-80: 39%


Comparison of Veto Techniques• Bug fix seems to decrease the performance of the “Extra

Electron Removal” Method– Due to the removal of unwanted tracks in QCD?

• Need to apply same rigorous selection to the extra electronsso an not to remove fake electrons in the QCD sample

DR < 0.3, with bug fixDR < 0.3, w/out bug fix


Efficiency curves with weighted QCD

• Overall performance has decreased due to the addition of the QCDPT_30-50 sample.

• Still seems as if there is no need to stray from inner cone method– Similar performance– Simplest implementation


Summary of Tracker Isolation

• Inner cone method is easiest to implement, and gives thebest isolation performance for signal versus background.

• For similar lepton purity, electron isolation performancealmost as good as muon isolation performance

34.6%98.3%ΣPT < 6 GeV

25.0%97.2%ΣPT < 4 GeV19.5%96.1%ΣPT < 3 GeV

29.8%97.9%ΣPT < 5 GeV

28.6%98.2%ΣPT < 4 GeV (Muons)*0.015 < ΔR < 0.3

14.6%93.7%ΣPT < 2 GeV

QCD IsolationEfficiency*

DY IsolationEfficiency*

CB LooseElectrons


Calo BasedIsolation


Previous Electron Footprint Removal in ECAL

• Sum the ET of Island Basic Clusters (IBCs) within a ΔR cone using thefollowing methods to veto the footprint of the electron:1. Subtract out the raw SC energy (SuperCluster Subtraction (SCS) Method)2. Remove IBCs within a smaller cone around the position of the electron where

it enters the calorimeter (the naïve approach) (Cone Method)3. Remove IBCs within a Δη strip around the position of the electron where it

enters the calorimeter (Strip Method)

SCS Strip Cone


Super Cluster Subtraction Method

• Compare isolation in the barrel and endcap– Fixed algorithm, now match HSC to a ISC and use ISC energy to subtract out

energy• Performance in the endcap unaffected, but better performance in the

barrel than previously.


New algo comparison to endcap

• Comparison of performance in the endcap to the barrel


New footprint removal technique - Jurassic

• New method to try in this pass.– Combination of both the cone and strip method.– Called Jurassic not because it was

implemented in ORCA, but because it isshaped like the sign in Jurassic Park

Strip Cone

Combine the cone and strip method to get the Jurassic Method

Jurassic

ΔR

ΔηΔRin


Optimizing Jurassic Method• Vary inner Δη strip, inner ΔR

cone, and outer ΔR cone• Choose:

– Δη > 0.01– 0.05 < ΔR < 0.4


ΔR < 0.4Cone: ΔR > 0.07Strip: Δη > 0.03

Jurassic: Δη > 0.01, ΔR > 0.05

ECAL electron isolation comparison

• Compare all four techniques with new QCD weighting.– New Jurassic method slightly outperforms the SC matching method,

and the strip method.– Cone method still lacks the ability to capture the electron footprint, and

hence performs worse.

43.5%97.1%ΣET < 4 GeV

35.1%95.5%ΣET < 3 GeV

51.4%97.9%ΣET < 5 GeV

*w/ Jurassic Method, Barrel

25.8%91.7%ΣET < 2 GeV

QCDIsolation

Efficiency*

DYIsolation

Efficiency*

CB LooseElectrons


Combined tracker and calorimeter performance

• Combine ECAL, HCAL and tracker Isolation– Should be a high correlation between the HCAL isolation and the

tracker isolation– See a slight performance increase by adding the HCAL

ΔR < 0.4Tracker: ΔR > 0.015

ECAL: Δη > 0.01, R < 0.05HCAL: ΔR > 0.0

17.0%97.1%ΣET < 10 GeV

11.9%95.8%ΣET < 8 GeV

22.6%97.9%ΣET < 12 GeV

*Barrel

7.45%93.2%ΣET < 6 GeV

QCDIsolation

Efficiency*

DYIsolation

Efficiency*

CB LooseElectrons


Summary and To Do

• Added new algorithm to ECAL isolation - Jurassic• Updated samples with full range of QCD samples

• Would like to add vertex compatibility• Need to optimize track quality cuts to better reflect current state of affairs

in tracking• Add electron ID to extra vetoed electrons within cone

Documents

A Update on Electron Isolation - WebHome < UCSDTier2 Update on Electron Isolation ... •Bug fix in Secondary Electron Algorithm ... DR < 0.3, w/out bug fix DR < 0.3, with bug fix